Predictive Performance of EV Charging Behaviour in COVID-19

Transportation is an essential prerequisite for modern life, however, the conventional combustion engine is rapidly losing its relevance. Vehicles powered by petrol and diesel emit a substantial amount of pollutants and are currently being rapidly replaced by fully electric vehicles (EVs). These EVs are emission-free and significantly more eco-friendly. The COVID-19 pandemic has had a significant impact on the global EVs market. The widespread adoption of EVs faces various challenges, including vehicle scheduling due to the limited EV charging facilities. Smart scheduling algorithms can address this issue by managing to improve EV scheduling by utilizing tools and algorithms fueled by data-driven approaches to predict EV charging demand. Previous research has utilized historical EV charging data to predict the behavior of departure time and energy needs. However, unexpected events like a pandemic can lead to changes in behavior, making predictions more challenging. This paper proposes using pre-COVID and COVID-period historical charging data to predict EV session duration using machine learning algorithms like Decision Tree Regressor, XGBoost, and Deep Neural Networks. XGBoost achieved the best predictive performance, with R2 scores of 0.99 for session duration, outperforming existing literature. The study demonstrates significant improvement in both predictions compared to previous work and conducts a comparative analysis of EV behavior during normal and unexpected circumstances. Overall as a result of the COVID-19 pandemic, the global EV market has experienced a decline in growth. With factory closures, supply chain disruptions, and reduced consumer demand, the production and sales of EVs have been negatively impacted.